Examples with ConnectorOutputMetadata io.trino.spi.connector.ConnectorOutputMetadata used on opensource projects

Example 1 with ConnectorOutputMetadata

use of io.trino.spi.connector.ConnectorOutputMetadata in project trino by trinodb.

the class HiveMetadata method finishInsert.

@Override
public Optional<ConnectorOutputMetadata> finishInsert(ConnectorSession session, ConnectorInsertTableHandle insertHandle, Collection<Slice> fragments, Collection<ComputedStatistics> computedStatistics) {
    HiveInsertTableHandle handle = (HiveInsertTableHandle) insertHandle;
    List<PartitionUpdate> partitionUpdates = fragments.stream().map(Slice::getBytes).map(partitionUpdateCodec::fromJson).collect(toImmutableList());
    HiveStorageFormat tableStorageFormat = handle.getTableStorageFormat();
    partitionUpdates = PartitionUpdate.mergePartitionUpdates(partitionUpdates);
    Table table = metastore.getTable(handle.getSchemaName(), handle.getTableName()).orElseThrow(() -> new TableNotFoundException(handle.getSchemaTableName()));
    if (!table.getStorage().getStorageFormat().getInputFormat().equals(tableStorageFormat.getInputFormat()) && isRespectTableFormat(session)) {
        throw new TrinoException(HIVE_CONCURRENT_MODIFICATION_DETECTED, "Table format changed during insert");
    }
    if (handle.getBucketProperty().isPresent() && isCreateEmptyBucketFiles(session)) {
        List<PartitionUpdate> partitionUpdatesForMissingBuckets = computePartitionUpdatesForMissingBuckets(session, handle, table, false, partitionUpdates);
        // replace partitionUpdates before creating the empty files so that those files will be cleaned up if we end up rollback
        partitionUpdates = PartitionUpdate.mergePartitionUpdates(concat(partitionUpdates, partitionUpdatesForMissingBuckets));
        for (PartitionUpdate partitionUpdate : partitionUpdatesForMissingBuckets) {
            Optional<Partition> partition = table.getPartitionColumns().isEmpty() ? Optional.empty() : Optional.of(buildPartitionObject(session, table, partitionUpdate));
            if (handle.isTransactional() && partition.isPresent()) {
                PartitionStatistics statistics = PartitionStatistics.builder().setBasicStatistics(partitionUpdate.getStatistics()).build();
                metastore.addPartition(session, handle.getSchemaName(), handle.getTableName(), partition.get(), partitionUpdate.getWritePath(), Optional.of(partitionUpdate.getFileNames()), statistics, handle.isRetriesEnabled());
            }
            createEmptyFiles(session, partitionUpdate.getWritePath(), table, partition, partitionUpdate.getFileNames());
        }
    }
    List<String> partitionedBy = table.getPartitionColumns().stream().map(Column::getName).collect(toImmutableList());
    Map<String, Type> columnTypes = handle.getInputColumns().stream().collect(toImmutableMap(HiveColumnHandle::getName, column -> column.getHiveType().getType(typeManager)));
    Map<List<String>, ComputedStatistics> partitionComputedStatistics = createComputedStatisticsToPartitionMap(computedStatistics, partitionedBy, columnTypes);
    for (PartitionUpdate partitionUpdate : partitionUpdates) {
        if (partitionUpdate.getName().isEmpty()) {
            // insert into unpartitioned table
            if (!table.getStorage().getStorageFormat().getInputFormat().equals(handle.getPartitionStorageFormat().getInputFormat()) && isRespectTableFormat(session)) {
                throw new TrinoException(HIVE_CONCURRENT_MODIFICATION_DETECTED, "Table format changed during insert");
            }
            PartitionStatistics partitionStatistics = createPartitionStatistics(partitionUpdate.getStatistics(), columnTypes, getColumnStatistics(partitionComputedStatistics, ImmutableList.of()));
            if (partitionUpdate.getUpdateMode() == OVERWRITE) {
                // get privileges from existing table
                PrincipalPrivileges principalPrivileges = fromHivePrivilegeInfos(metastore.listTablePrivileges(handle.getSchemaName(), handle.getTableName(), Optional.empty()));
                // first drop it
                metastore.dropTable(session, handle.getSchemaName(), handle.getTableName());
                // create the table with the new location
                metastore.createTable(session, table, principalPrivileges, Optional.of(partitionUpdate.getWritePath()), Optional.of(partitionUpdate.getFileNames()), false, partitionStatistics, handle.isRetriesEnabled());
            } else if (partitionUpdate.getUpdateMode() == NEW || partitionUpdate.getUpdateMode() == APPEND) {
                // insert into unpartitioned table
                metastore.finishInsertIntoExistingTable(session, handle.getSchemaName(), handle.getTableName(), partitionUpdate.getWritePath(), partitionUpdate.getFileNames(), partitionStatistics, handle.isRetriesEnabled());
            } else {
                throw new IllegalArgumentException("Unsupported update mode: " + partitionUpdate.getUpdateMode());
            }
        } else if (partitionUpdate.getUpdateMode() == APPEND) {
            // insert into existing partition
            List<String> partitionValues = toPartitionValues(partitionUpdate.getName());
            PartitionStatistics partitionStatistics = createPartitionStatistics(partitionUpdate.getStatistics(), columnTypes, getColumnStatistics(partitionComputedStatistics, partitionValues));
            metastore.finishInsertIntoExistingPartition(session, handle.getSchemaName(), handle.getTableName(), partitionValues, partitionUpdate.getWritePath(), partitionUpdate.getFileNames(), partitionStatistics, handle.isRetriesEnabled());
        } else if (partitionUpdate.getUpdateMode() == NEW || partitionUpdate.getUpdateMode() == OVERWRITE) {
            // insert into new partition or overwrite existing partition
            Partition partition = buildPartitionObject(session, table, partitionUpdate);
            if (!partition.getStorage().getStorageFormat().getInputFormat().equals(handle.getPartitionStorageFormat().getInputFormat()) && isRespectTableFormat(session)) {
                throw new TrinoException(HIVE_CONCURRENT_MODIFICATION_DETECTED, "Partition format changed during insert");
            }
            PartitionStatistics partitionStatistics = createPartitionStatistics(partitionUpdate.getStatistics(), columnTypes, getColumnStatistics(partitionComputedStatistics, partition.getValues()));
            if (partitionUpdate.getUpdateMode() == OVERWRITE) {
                if (handle.getLocationHandle().getWriteMode() == DIRECT_TO_TARGET_EXISTING_DIRECTORY) {
                    removeNonCurrentQueryFiles(session, partitionUpdate.getTargetPath());
                    if (handle.isRetriesEnabled()) {
                        HdfsContext hdfsContext = new HdfsContext(session);
                        cleanExtraOutputFiles(hdfsEnvironment, hdfsContext, session.getQueryId(), partitionUpdate.getTargetPath(), ImmutableSet.copyOf(partitionUpdate.getFileNames()));
                    }
                } else {
                    metastore.dropPartition(session, handle.getSchemaName(), handle.getTableName(), partition.getValues(), true);
                    metastore.addPartition(session, handle.getSchemaName(), handle.getTableName(), partition, partitionUpdate.getWritePath(), Optional.of(partitionUpdate.getFileNames()), partitionStatistics, handle.isRetriesEnabled());
                }
            } else {
                metastore.addPartition(session, handle.getSchemaName(), handle.getTableName(), partition, partitionUpdate.getWritePath(), Optional.of(partitionUpdate.getFileNames()), partitionStatistics, handle.isRetriesEnabled());
            }
        } else {
            throw new IllegalArgumentException(format("Unsupported update mode: %s", partitionUpdate.getUpdateMode()));
        }
    }
    if (isFullAcidTable(table.getParameters())) {
        HdfsContext context = new HdfsContext(session);
        for (PartitionUpdate update : partitionUpdates) {
            long writeId = handle.getTransaction().getWriteId();
            Path deltaDirectory = new Path(format("%s/%s/%s", table.getStorage().getLocation(), update.getName(), deltaSubdir(writeId, writeId, 0)));
            createOrcAcidVersionFile(context, deltaDirectory);
        }
    }
    return Optional.of(new HiveWrittenPartitions(partitionUpdates.stream().map(PartitionUpdate::getName).collect(toImmutableList())));
}

Example 2 with ConnectorOutputMetadata

use of io.trino.spi.connector.ConnectorOutputMetadata in project trino by trinodb.

the class IcebergMetadata method finishInsert.

@Override
public Optional<ConnectorOutputMetadata> finishInsert(ConnectorSession session, ConnectorInsertTableHandle insertHandle, Collection<Slice> fragments, Collection<ComputedStatistics> computedStatistics) {
    IcebergWritableTableHandle table = (IcebergWritableTableHandle) insertHandle;
    Table icebergTable = transaction.table();
    List<CommitTaskData> commitTasks = fragments.stream().map(slice -> commitTaskCodec.fromJson(slice.getBytes())).collect(toImmutableList());
    Type[] partitionColumnTypes = icebergTable.spec().fields().stream().map(field -> field.transform().getResultType(icebergTable.schema().findType(field.sourceId()))).toArray(Type[]::new);
    AppendFiles appendFiles = transaction.newAppend();
    ImmutableSet.Builder<String> writtenFiles = ImmutableSet.builder();
    for (CommitTaskData task : commitTasks) {
        DataFiles.Builder builder = DataFiles.builder(icebergTable.spec()).withPath(task.getPath()).withFileSizeInBytes(task.getFileSizeInBytes()).withFormat(table.getFileFormat().toIceberg()).withMetrics(task.getMetrics().metrics());
        if (!icebergTable.spec().fields().isEmpty()) {
            String partitionDataJson = task.getPartitionDataJson().orElseThrow(() -> new VerifyException("No partition data for partitioned table"));
            builder.withPartition(PartitionData.fromJson(partitionDataJson, partitionColumnTypes));
        }
        appendFiles.appendFile(builder.build());
        writtenFiles.add(task.getPath());
    }
    // try to leave as little garbage as possible behind
    if (table.getRetryMode() != NO_RETRIES) {
        cleanExtraOutputFiles(session, writtenFiles.build());
    }
    appendFiles.commit();
    transaction.commitTransaction();
    transaction = null;
    return Optional.of(new HiveWrittenPartitions(commitTasks.stream().map(CommitTaskData::getPath).collect(toImmutableList())));
}

Example 3 with ConnectorOutputMetadata

use of io.trino.spi.connector.ConnectorOutputMetadata in project trino by trinodb.

the class IcebergMetadata method finishRefreshMaterializedView.

@Override
public Optional<ConnectorOutputMetadata> finishRefreshMaterializedView(ConnectorSession session, ConnectorTableHandle tableHandle, ConnectorInsertTableHandle insertHandle, Collection<Slice> fragments, Collection<ComputedStatistics> computedStatistics, List<ConnectorTableHandle> sourceTableHandles) {
    // delete before insert .. simulating overwrite
    executeDelete(session, tableHandle);
    IcebergWritableTableHandle table = (IcebergWritableTableHandle) insertHandle;
    Table icebergTable = transaction.table();
    List<CommitTaskData> commitTasks = fragments.stream().map(slice -> commitTaskCodec.fromJson(slice.getBytes())).collect(toImmutableList());
    Type[] partitionColumnTypes = icebergTable.spec().fields().stream().map(field -> field.transform().getResultType(icebergTable.schema().findType(field.sourceId()))).toArray(Type[]::new);
    AppendFiles appendFiles = transaction.newFastAppend();
    ImmutableSet.Builder<String> writtenFiles = ImmutableSet.builder();
    for (CommitTaskData task : commitTasks) {
        DataFiles.Builder builder = DataFiles.builder(icebergTable.spec()).withPath(task.getPath()).withFileSizeInBytes(task.getFileSizeInBytes()).withFormat(table.getFileFormat().toIceberg()).withMetrics(task.getMetrics().metrics());
        if (!icebergTable.spec().fields().isEmpty()) {
            String partitionDataJson = task.getPartitionDataJson().orElseThrow(() -> new VerifyException("No partition data for partitioned table"));
            builder.withPartition(PartitionData.fromJson(partitionDataJson, partitionColumnTypes));
        }
        appendFiles.appendFile(builder.build());
        writtenFiles.add(task.getPath());
    }
    String dependencies = sourceTableHandles.stream().map(handle -> (IcebergTableHandle) handle).filter(handle -> handle.getSnapshotId().isPresent()).map(handle -> handle.getSchemaTableName() + "=" + handle.getSnapshotId().get()).distinct().collect(joining(","));
    // try to leave as little garbage as possible behind
    if (table.getRetryMode() != NO_RETRIES) {
        cleanExtraOutputFiles(session, writtenFiles.build());
    }
    // Update the 'dependsOnTables' property that tracks tables on which the materialized view depends and the corresponding snapshot ids of the tables
    appendFiles.set(DEPENDS_ON_TABLES, dependencies);
    appendFiles.commit();
    transaction.commitTransaction();
    transaction = null;
    return Optional.of(new HiveWrittenPartitions(commitTasks.stream().map(CommitTaskData::getPath).collect(toImmutableList())));
}

Example 4 with ConnectorOutputMetadata

use of io.trino.spi.connector.ConnectorOutputMetadata in project trino by trinodb.

the class HiveMetadata method finishCreateTable.

@Override
public Optional<ConnectorOutputMetadata> finishCreateTable(ConnectorSession session, ConnectorOutputTableHandle tableHandle, Collection<Slice> fragments, Collection<ComputedStatistics> computedStatistics) {
    HiveOutputTableHandle handle = (HiveOutputTableHandle) tableHandle;
    List<PartitionUpdate> partitionUpdates = fragments.stream().map(Slice::getBytes).map(partitionUpdateCodec::fromJson).collect(toImmutableList());
    WriteInfo writeInfo = locationService.getQueryWriteInfo(handle.getLocationHandle());
    Table table = buildTableObject(session.getQueryId(), handle.getSchemaName(), handle.getTableName(), handle.getTableOwner(), handle.getInputColumns(), handle.getTableStorageFormat(), handle.getPartitionedBy(), handle.getBucketProperty(), handle.getAdditionalTableParameters(), Optional.of(writeInfo.getTargetPath()), handle.isExternal(), prestoVersion, accessControlMetadata.isUsingSystemSecurity());
    PrincipalPrivileges principalPrivileges = accessControlMetadata.isUsingSystemSecurity() ? NO_PRIVILEGES : buildInitialPrivilegeSet(handle.getTableOwner());
    partitionUpdates = PartitionUpdate.mergePartitionUpdates(partitionUpdates);
    if (handle.getBucketProperty().isPresent() && isCreateEmptyBucketFiles(session)) {
        List<PartitionUpdate> partitionUpdatesForMissingBuckets = computePartitionUpdatesForMissingBuckets(session, handle, table, true, partitionUpdates);
        // replace partitionUpdates before creating the empty files so that those files will be cleaned up if we end up rollback
        partitionUpdates = PartitionUpdate.mergePartitionUpdates(concat(partitionUpdates, partitionUpdatesForMissingBuckets));
        for (PartitionUpdate partitionUpdate : partitionUpdatesForMissingBuckets) {
            Optional<Partition> partition = table.getPartitionColumns().isEmpty() ? Optional.empty() : Optional.of(buildPartitionObject(session, table, partitionUpdate));
            createEmptyFiles(session, partitionUpdate.getWritePath(), table, partition, partitionUpdate.getFileNames());
        }
        if (handle.isTransactional()) {
            AcidTransaction transaction = handle.getTransaction();
            List<String> partitionNames = partitionUpdates.stream().map(PartitionUpdate::getName).collect(toImmutableList());
            metastore.addDynamicPartitions(handle.getSchemaName(), handle.getTableName(), partitionNames, transaction.getAcidTransactionId(), transaction.getWriteId(), AcidOperation.CREATE_TABLE);
        }
    }
    Map<String, Type> columnTypes = handle.getInputColumns().stream().collect(toImmutableMap(HiveColumnHandle::getName, column -> column.getHiveType().getType(typeManager)));
    Map<List<String>, ComputedStatistics> partitionComputedStatistics = createComputedStatisticsToPartitionMap(computedStatistics, handle.getPartitionedBy(), columnTypes);
    PartitionStatistics tableStatistics;
    if (table.getPartitionColumns().isEmpty()) {
        HiveBasicStatistics basicStatistics = partitionUpdates.stream().map(PartitionUpdate::getStatistics).reduce((first, second) -> reduce(first, second, ADD)).orElse(createZeroStatistics());
        tableStatistics = createPartitionStatistics(basicStatistics, columnTypes, getColumnStatistics(partitionComputedStatistics, ImmutableList.of()));
    } else {
        tableStatistics = new PartitionStatistics(createEmptyStatistics(), ImmutableMap.of());
    }
    if (handle.getPartitionedBy().isEmpty()) {
        List<String> fileNames;
        if (partitionUpdates.isEmpty()) {
            // creating empty table via CTAS ... WITH NO DATA
            fileNames = ImmutableList.of();
        } else {
            fileNames = getOnlyElement(partitionUpdates).getFileNames();
        }
        metastore.createTable(session, table, principalPrivileges, Optional.of(writeInfo.getWritePath()), Optional.of(fileNames), false, tableStatistics, handle.isRetriesEnabled());
    } else {
        metastore.createTable(session, table, principalPrivileges, Optional.of(writeInfo.getWritePath()), Optional.empty(), false, tableStatistics, false);
    }
    if (!handle.getPartitionedBy().isEmpty()) {
        if (isRespectTableFormat(session)) {
            verify(handle.getPartitionStorageFormat() == handle.getTableStorageFormat());
        }
        for (PartitionUpdate update : partitionUpdates) {
            Partition partition = buildPartitionObject(session, table, update);
            PartitionStatistics partitionStatistics = createPartitionStatistics(update.getStatistics(), columnTypes, getColumnStatistics(partitionComputedStatistics, partition.getValues()));
            metastore.addPartition(session, handle.getSchemaName(), handle.getTableName(), buildPartitionObject(session, table, update), update.getWritePath(), Optional.of(update.getFileNames()), partitionStatistics, handle.isRetriesEnabled());
        }
    }
    return Optional.of(new HiveWrittenPartitions(partitionUpdates.stream().map(PartitionUpdate::getName).collect(toImmutableList())));
}